Reasoning
EmbodiedBench
MLLMs act as vision-driven embodied agents across four simulated environments; response is per-episode binary task success.
969items
11subjects
MITlicense
agents_and_tool_usedomain
reasoningdomain
textmodality
imagemodality
item-level responses released
Saturation status: Unknown
Response matrix
Fit to width. Hover for subject & item; click a cell for details.
Correct (1)Incorrect (0)Unobserved
Scale: 1 = correct · 0 = incorrect
Subjects
- 1Claude 3.7 Sonnet0.5522
- 2Claude 3.5 Sonnet0.5121
- 3GPT-4o0.4832
- 4InternVL3-78B0.4556
- 5Gemini 1.5 Pro0.4149
- 6Qwen-VL-Max0.3843
- 7Gemini 2.0 Flash0.3792
- 8Ovis2-34B0.3767
- 9Qwen2.5-VL-72B-Instruct0.3594
- 10InternVL2.5-78B0.3449
- 11Llama-3.2-90B-Vision-Instruct0.2805